Evaluation of a clustering algorithm for texture data

In forming simulations of complex part designs, material texture can play a crucial role. However, spatially resolved integration of texture is challenging due to large data size. A reduction in data size can be achieved by meso-scale approaches, such as the viscoplastic self-consistent (VPSC) model. The VPSC model calculates individual grain responses within a deformed matrix, therefore the total number of grains has a substantial impact on the computation time. In this work, an algorithm is presented that cumulatively reduces the number of grains, without causing significant deviations in the simulation results. Our approach is based on a k-means algorithm. Instead of setting the number of k clusters, a fixed radius is used. The size of this cluster radius determines the degree of data reduction. The impact of clustering-induced errors is evaluated for an extruded EN AW-6082 alloy via texture investigations and the flow curves of simulated tensile tests. These simulations were performed using the VPSC approach as well as a finite element model in combination with VPSC. The results provide an upper limit for data reduction with the presented algorithm. • Orientations of an EN AW-6082 texture were reduced via a clustering algorithm. • The algorithm performance is validated via texture analysis and VPSC simulations. • Simulations of tensile testing were performed via VPSC Standalone and FE-VPSC. • By clustering the texture orientations VPSC calculation times can be reduced.